LoRASculpt: Sculpting LoRA for Harmonizing General and Specialized Knowledge in Multimodal Large Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Liang, Jian, Huang, Wenke, Wan, Guancheng, Yang, Qu, Ye, Mang
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915207911047168
author Liang, Jian
Huang, Wenke
Wan, Guancheng
Yang, Qu
Ye, Mang
author_facet Liang, Jian
Huang, Wenke
Wan, Guancheng
Yang, Qu
Ye, Mang
contents While Multimodal Large Language Models (MLLMs) excel at generalizing across modalities and tasks, effectively adapting them to specific downstream tasks while simultaneously retaining both general and specialized knowledge remains challenging. Although Low-Rank Adaptation (LoRA) is widely used to efficiently acquire specialized knowledge in MLLMs, it introduces substantial harmful redundancy during visual instruction tuning, which exacerbates the forgetting of general knowledge and degrades downstream task performance. To address this issue, we propose LoRASculpt to eliminate harmful redundant parameters, thereby harmonizing general and specialized knowledge. Specifically, under theoretical guarantees, we introduce sparse updates into LoRA to discard redundant parameters effectively. Furthermore, we propose a Conflict Mitigation Regularizer to refine the update trajectory of LoRA, mitigating knowledge conflicts with the pretrained weights. Extensive experimental results demonstrate that even at very high degree of sparsity ($\le$ 5%), our method simultaneously enhances generalization and downstream task performance. This confirms that our approach effectively mitigates the catastrophic forgetting issue and further promotes knowledge harmonization in MLLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2503_16843
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LoRASculpt: Sculpting LoRA for Harmonizing General and Specialized Knowledge in Multimodal Large Language Models
Liang, Jian
Huang, Wenke
Wan, Guancheng
Yang, Qu
Ye, Mang
Computer Vision and Pattern Recognition
While Multimodal Large Language Models (MLLMs) excel at generalizing across modalities and tasks, effectively adapting them to specific downstream tasks while simultaneously retaining both general and specialized knowledge remains challenging. Although Low-Rank Adaptation (LoRA) is widely used to efficiently acquire specialized knowledge in MLLMs, it introduces substantial harmful redundancy during visual instruction tuning, which exacerbates the forgetting of general knowledge and degrades downstream task performance. To address this issue, we propose LoRASculpt to eliminate harmful redundant parameters, thereby harmonizing general and specialized knowledge. Specifically, under theoretical guarantees, we introduce sparse updates into LoRA to discard redundant parameters effectively. Furthermore, we propose a Conflict Mitigation Regularizer to refine the update trajectory of LoRA, mitigating knowledge conflicts with the pretrained weights. Extensive experimental results demonstrate that even at very high degree of sparsity ($\le$ 5%), our method simultaneously enhances generalization and downstream task performance. This confirms that our approach effectively mitigates the catastrophic forgetting issue and further promotes knowledge harmonization in MLLMs.
title LoRASculpt: Sculpting LoRA for Harmonizing General and Specialized Knowledge in Multimodal Large Language Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2503.16843